Summary Semblance measures similarity among short-time signals along a trajectory spread over a line or area in space. Seismic processors use it for velocity analysis. Seismic interpreters use it to identify faults, channels and other stratigraphic features. However, the effectiveness of conventional time domain semblance algorithms is limited. These algorithms require long temporal windows for a robust measure of similarity. This long time-window limits the ability of the method to localize the signal in time which is typically one or one-half period of a reflection signal. Alternatively, the use of a short time-window makes the semblance more susceptible to effects of local noise and introduces a dc bias. Our proposed scheme provides an additional degree of freedom in the form of time-frequency volumes. We improve the fidelity of our semblance volume while using even very short time-windows. We also propose a two-step approach to counter the effect of noisy frequencies by intelligent pass-band selection and amplitude weighted averaging of semblances computed in this pass-band. This enables tuned feature detection. We demonstrate the effectiveness of the method with sharp identification of boundaries of a channel that lies on a dipping 3D structure.